基于模糊模式识别的光谱分类研究

【作者】 罗磊；

【导师】 赵元黎；

【作者基本信息】 郑州大学 ， 测试计量技术与仪器， 2006， 硕士

【摘要】 模式识别是研究分类识别理论和方法的科学，在理论上涉及到代数学、矩阵论、概率论、图论、模糊数学、最优化理论等等众多学科的知识，其内涵可概括为信息处理、分析、决策。其一般过程包括模式信息采集、预处理、特征提取与选择、模式分类这几个步骤；其方法主要包括统计模式识别、句法模式识别、模糊数学方法、神经网络法、人工智能方法等，针对不同的对象和不同的目的，而选用不同的模式识别理论、方法。模式识别的理论、技术已成功地应用于工业、农业、国防、科研、公安、生物医学、气象、天文学等诸多领域。 光谱技术是通过光和物质的相互作用来了解物质结构及其性质的一种方法。随着激光的问世和计算机的普及发展，各种光学、光谱仪器也得到迅速发展，仪器的精度有很大提高，能够采集到非常微弱的光电信号，应用光谱技术检测医学样品、生物样品，进行医学诊断成为可能。 本文分析比较了各种模式识别方法的优缺点，依据待处理样品的光谱本身具有模糊性的特点，选择了模糊模式识别方法，分别对乳腺肿瘤周边组织拉曼光谱和各类血浆样品的紫外吸收光谱进行了处理。 在乳腺肿瘤周边组织拉曼光谱数据的预处理中，对FFT滤波和临域平均值法两种不同的平滑方法在光谱平滑处理中的效果进行了比较分析，并比较了单次与多次平滑的效果以及对后续工作的影响，选取了对光谱处理更适用的临域平均值法对光谱进行了一次平滑处理；观察分析了大量谱线，提取了光谱的多个特征进行分析比较，并选择了符合率较高的I₁₄₄₀／I₁₅₀₁和1440cm^-1与1530cm^-1波段拉曼峰连线斜率的绝对值作为分类识别的特征参数；对选择的特征参数采用秩和检验法进行了统计学检验，证明了所选参数的有效性；改进了梯形分布的偏大型隶属函数，提出了分段式多维偏大型隶属函数PMBMF（piecewise multidimensional leaning towards the bigger membership function）作为对谱线进行模糊模式识别的隶属函数，利用PMBMF设计了符合率较高的分类器，该分类器对癌症病例的识别率为82.4％，为非癌病例的识别率为73.9％；对分类更多还原

【Abstract】 Pattern recognition theory comes down to algebra, matrix theory, probability theory, graph theory, fuzzing mathematics and optimization theory.Its meaning can be summed up by information processing, analysis and deciding. Its normal process generally includes information collection, preconditioning, feature extraction and selection and pattern classifying.Its methods contain statistical pattern recognition, syntax pattern recognition, fuzzing mathematics technique,NN technique and artificial intelligence, etc. Difference theory or method will be selected when people aim at various different purposes or object. Pattern recognition theory and technoloy has succeeded in many fiels which include industy, agriculture, national defence, scientific research, public security, biomedicine, weather and astronomy and so on.Spectrum technoloy can find substance structure and character through the interaction of light and substance.All kinds of optical apparatuses and spectrum apparatuses have been developed rapidly with the invention of laser and the development of computer. What’ s more , the precision of various apparatuses have been improved greatly and many weak photoelectric signals can be collected. Then , it is possible that spectrum technoloy is applied in diagnosis.This article analyzes and compares the advantage and disadvantage of all kinds of pattern recognition methods. fuzzing pattern recognition was selected to process various Ramam spectra of tumor peripheral tissue and ultraviolet absorbing spectra of plasma based on the samples’ fuzzyness.We comapared the smoothing effects which come from FFT filtering and adjacent averaging then selected the adjacent averaging as the method to更多还原